OBJECTIVES: To assess the relative ability of several health-related quality of life (HRQOL) measures to predict future health care expenditures.  METHODS:  Data from the Medical Expenditure Panel Survey (MEPS) for years 2003 (Panel 4, Round 2) and 2004 were used for study purposes and weighted in order to gain a population-level analysis. Applying the Model of Health Services Use, predisposing (age, gender, race, years of education, and marital status) and enabling (insurance type, employment status, family size, and annual household income) variables were combined with varying need variables (SF-12 PCS, SF-12 MCS, EQ-5D Index, EQ-5D VAS, single-item core questions of perceived physical and mental health status, or a combination thereof) and used to predict overall healthcare expenditures, one year after survey, in multivariate linear regression models. The individual R2 values were used for model comparison. RESULTS:  The final dataset was composed of 9304 respondents, representing over 186 million US residents. The base model of only predisposing and enabling covariates resulted in an R2 value of 0.067. The model using both the SF-12 PCS and MCS values as need variables resulted in the highest R2 value of all models run: 0.094. Use of the SF-12 MCS or the single-item perceived mental health core item as need variables only marginally out-performed the base model, both resulting in R2 values of 0.069. The EQ-5D VAS and Index as well as the physical health perception core measure values were similar, ranging from 0.0821 to 0.0868. CONCLUSIONS:  The combined use of the SF-12 MCS and PCS measures as need variables in the Model of Health Services Use performed better than the other HRQOL measures in the MEPS dataset in predicting future health care expenditures.